[Note: Unzip the download to get the GGUF. Civit doesn't support it natively, hence this workaround]
A merge of FluxUnchained and HyperSD LoRA from ByteDance - converted to GGUF. As a result, it can now generate artistic NSFW images in 8 steps while consuming very low VRAM. The Q_4_0 model consumes around 6.5 GB VRAM and takes around 1.5 min to generate a 1024x1024 image with 8 steps on my 1080ti. [See https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050 to learn more about Forge UI GGUF support and also where to download the VAE, clip_l and t5xxl models.]
You can also combine it with other LoRAs to get the effect you want.
Advantages Over FastFlux Unchained and HyFU
Much better quality: you get much better quality and expressiveness at 8 steps compared to Schnell models like HyFU
CFG/Guidance Sensitivity: Since this is a DEV model, unlike the Hybrid models, you get full (distilled) CFG sensitivity - i.e., you can control prompt sensitivity vs. creativity and softness vs. saturation.
Fully compatible with Dev LoRAs, better than the compatibility of Schnell models.
The only disadvantage: needs 8-step for best quality. But then, you'd probably try at least 8 steps for best results with Schnell anyway.
Which model should I download?
[Current situation: Using the updated Forge UI and Comfy UI (GGUF node) I can run Q8_0 on my 11GB 1080ti.]
Download the one that fits in your VRAM. The additional inference cost is quite small if the model fits in the GPU. Size order is Q4_0 < Q4_1 < Q5_0 < Q5_1 < Q8_0.
Q4_0 and Q4_1 should fit in 8 GB VRAM
Q5_0 and Q5_1 should fit in 11 GB VRAM
Q8_0 if you have more!
Note: With CPU offloading, you will be able to run a model even if doesn't fit in your VRAM.
All the license terms associated with Flux.1 Dev apply.
PS: Credit goes to ByteDance for the HyperSD Flux 8-steps LoRA which can be found at https://huggingface.co/ByteDance/Hyper-SD/tree/main
Description
FAQ
Comments (12)
adding the GGUF to comfyUI isn't difficult, the quality is okay but it is fast with the compared to base flux, age of men and women seem to be set like on/off of 25 or 80
Terrible model, all images are covered in noise and if you use lora everything turns into a mess of noise. Lost 100 buzzes on this horror.
I'm sorry to hear that it didn't work for you. I'll refund you those buzzes.
@nakif0968 Why does the model make all images noisy, is it because of the hyper? If connect the lora, the images completely turn into a mess of noise.
@SuperSmuser You don't need the LoRA, it has been merged into the model.
@nakif0968 I didn't use hyper lora, I mean any lora. All the loras I connected turn the images into mush.
I don't get any noise but I have found all GGUF models are a bit finicky in ComfyUI. I've been having a problem with high contrast images in pretty much every model I've tried even just from adding an extra detail to a prompt.
@SuperSmuser I use comfyui, I had to use the FluxGuidance node and set it to at least 3.5 and use euler with beta scheduler. Before this I could only use schnell models.
Works fine here? see review i added, don't get any noise on my images (dunno about with lora's but definitely not without)
@baconmessenger You have noise in your images, but it's a different type of noise, not the one I was talking about, but it's also terrible. Try generating asphalt and it will be covered with the pattern that covers all the images. If you still can't see it, I can try to point it out.
But it's not that scary, what's scarier are the artifacts that cover the images, especially with lora. Try connecting any strong lora and just generate "space" and you'll get an image as if it were in 144p. And this quality is almost the same with any prompts, even without lora.
how many steps?
man this really shouldn't have been posted as a .zip archive, civitai has issues disconnecting mid-download, and archives have to start all over every time. impossible download lol